Incorporating landscape context into species distribution models improves predictions for migratory shorebirds

Dataset for: Incorporating landscape context into species distribution models improves predictions for migratory shorebirds This dataset supports the findings reported in [Incorporating landscape context into species distribution models improves predictions for migratory shorebirds]. The study used a nested species distribution modeling (N-SDM) framework to evaluate the contribution of landscape-scale variables to habitat suitability predictions for six migratory shorebird species in the East Asian-Australasian Flyway (EAAF) during the non-breeding season. The framework combines a climate layer fitted across the entire EAAF (1-km resolution) with a coastal habitat layer fitted within the coastal zone (500-m resolution). Contents: Occurrence records (CSV): Spatially filtered eBird occurrence records for six shorebird species. Records were drawn from complete checklists meeting strict effort criteria (duration ≤ 6 hours, travel distance ≤ 5 km, and ≤ 10 observers) during the core non-breeding period (December–February, 2021–2023). To reduce local sampling redundancy while preserving fine-scale information, one record was randomly selected from each occupied 1 km × 1 km grid cell for each species, matching the coarsest model resolution. Species abbreviations: batgod, Bar-tailed Godwit (Limosa lapponica); bkbplo, Black-bellied Plover (Pluvialis squatarola); comgre, Common Greenshank (Tringa nebularia); eurcur, Eurasian Curlew (Numenius arquata); grekno, Great Knot (Calidris tenuirostris); grsplo, Greater Sand-Plover (Anarhynchus leschenaultii). Predicted habitat suitability maps (GeoTIFF): Relative habitat suitability maps at 500-m resolution for each species, produced from the combined climate and coastal habitat layers of the N-SDM framework. Maps cover the EAAF coastal zone (60°N–60°S). Modeling code (Python): A Random Forest species distribution modeling script implemented in Google Earth Engine, covering predictor assembly, spatially blocked cross-validation, model training and prediction, evaluation (AUC-ROC and AUC-PR), and export of habitat suitability maps. Adapted from Crego et al. (2022); raw input data are archived on Google Earth Engine, with asset paths replaced by descriptive placeholders.